Primary Storage

Trip report ‘IBM Spectrum Scale / Elastic Storage / GPFS Expert Workshop Ehningen 2015' (Part I)

By Archive User posted Fri March 06, 2015 04:09 AM


Originally posted by: Ulf Troppens

The ‘IBM Spectrum Scale / Elastic Storage / GPFS Expert Workshop 2015’ was a 2-day workshop delivered by experts for experts. This year we had again a mix of speakers ranging from customers and business partner to IBM technical services, development and research.

Before the official start Frank Krämer warmed up at 8.30am with ‘IBM Spectrum Scale for Beginners’. I was not able to join, because I had a customer briefing in parallel, but I was told that this session was very well attended. I like Frank’s introduction where he compares file systems and in particular scalable file systems with red wine. Both need time to mature and to develop its character. IBM started Tiger Shark as research project in the early 1990ies which evolved to GPFS and Elastic Storage. The new name is IBM Spectrum Scale which inherits a large install base from GPFS, guaranteeing a revenue stream to fund future enhancements. This is a huge advantage compared to open source file system which do have less maturity and a less solid funding stream for future development. For getting started with IBM Spectrum Scale (a.k.a Elastic Storage (a.k.a GPFS)) I recommend ‘Software Defined Storage for Dummies’.

At 10.00am Götz Mensel opened the official program. He excused that this year the workshop cannot be in Mainz – the head quarter of IBM storage in Germany. After 110 attendees in 2013 and 180 attendees last year, we have 250 attendees in 2015. The facilities is Mainz are just not capable to host such a large group. Götz did a great job to moderate the two days.

Dieter Münk explained the big picture of ‘Infrastructure Matters’. He said that being this year in Ehningen because of the large audience is the best proof that Infrastructures Matters. He explained that the integration of System of Records (=traditional IT) and Systems of Engagements (=new IT) enables new Systems of Insight (=new insight). He recently visited a customer who is analyzing Twitter tweets (=Systems of Engagement) to monitor the health status (=insight) of their ATMs (=System of Records). Dieter said that IBM is fully committed to infrastructure. IBM will continue to deliver hardware, where we can make a difference to the market such as IBM z Systems (mainframe) and IBM Elastic Storage Server. In other areas we will focus on software. See Tony Person’s Inside System Storage blog for ‘How is Software Defined different than what we have now?

Frank Bröde continued with an overview of the new IBM Spectrum Storage family. IBM just announced to invest 1 Billion US-$ in Software Defined Storage. IBM is not starting from scratch. IBM Spectrum Storage is based on existing successful IBM storage hardware and software products. This investment is used to accelerate the transition to new Software Defined world. Tony Person has also a good blog on how everything fits together: The Software Defined Quadrant.

Alexander Gast reported about the use of GPFS at Westdeutscher Rundfunk (WDR), the second largest broadcaster in Europe. WDR started in 2012 to use IBM AREMA to support the whole workflow from importing the media via transcoding and post production to delivery. GPFS helps them to collaborate more effectively and to process the whole workflow more quickly. They are telling that GPFS provides high performance and is robust and high available. One nice side effect is that having GPFS as shared back-end storage reduces the copy steps from five to two saving time and storage capacity.

Next Alexander Saupp did a dive deep into IBM Elastic Storage Server. The continuous need to process more and more amounts of storage requires to operate large file systems comprising huge amounts of large drives for mission critical workload. Failing drives is a frequent event due to the huge amounts of drives. RAID rebuilds take longer due to the size of the drives. In addition, hitting a non-recoverable bit error during rebuild is more likely, given the huge amounts of data which need to be read during traditional RAID rebuild. IBM invented the GPFS Native RAID (GNR) (video with explanations) to cope with these challenges. GNR is integrated in IBM Elastic Storage Server. At the end of his session Alexander briefly touched proven blueprints which integrates IBM Elastic Storage Server. I skip them here, because they have been covered in more details in later sessions.

Andre Gaschler was the last speaker before lunch. He reported about three proof-of-concepts with Elastic Storage Server as TSM Storage Pool boosting the performance of backup and restore. They investigated setups with 56 Gigabit InfiniBand, 40 Gigabit Ethernet and 10 Gigabit Ethernet and have achieved unparalleled results. For all tests they have used a very simple setup. They just created one large GPFS file system with a few subdirectories for separation. See here for details of the proof-of-concepts with InfiniBand and 40GBE. The results for 10GBE will be published as TSM Blueprint soon.